This study’s objective was to evaluate the CGM’s sensitivity to marker placement. We simulated different combinations of markers on the lower-left limb, displaced in different directions by 10 mm. Overall, measurements in the transversal plane demonstrated the greatest sensitivity to marker displacement, whereas the markers displaced in the sagittal plane resulted in the highest RMSD angles in comparison to the original kinematics.
Pelvic kinematics showed very low sensitivity to marker displacement, with all their RMSD angles calculated to within an acceptable limit of 5° and the majority of their simulations resulting in RMSD angles within 2°. For the other joint angles calculated, the transversal plane was the most affected by marker displacement, with about 47% of simulations returning an error over the 5° limit of acceptability. These findings agreed with previous literature reporting that the transversal plane was the least reliable in gait analysis2,16.
An analysis of the ten worst-case marker misplacement scenarios allowed us to better understand the effects of a combination of marker displacements on the lower-limb model. For instance, the ‘worst’ marker configuration for the pelvis was calculated for when the anterior iliac spines markers were displaced in opposite directions in the vertical axis and the SACR was displaced in the horizontal axis. With this simulated marker configuration, the pelvis was both tilted and rotated with respect to its original definition. Because the CGM is a hierarchical, anatomical, top-down model, this would be expected to affect the hip-joint centre estimation, hip kinematics, and all the distal joint angles. As illustrated in Figure 3, the CGM’s thigh and tibia’s flexion–extension axes are defined as orthogonal to the plane connecting the proximal and distal joint centres when the wand is placed along the segment. Thus, those segments’ medial–lateral axes are estimated to be orthogonal to both the flexion–extension and proximal–distal axes. The simulated displacement of the femoral wand (LTHI) in the anterior–posterior axis directly affects the femur’s coronal plane, thus altering the flexion–extension axis and the medial–lateral axis. As a consequence, the kinematics of the hip and knee joints will be directly affected, as will the knee joint centre that is defined along the femur’s medial–lateral axis (in the absence of the medial femoral epicondyle marker). A similar impact was noted for the tibia. Finally, the medial displacement of the LTOE marker was responsible for a rotation of the foot’s angle with respect to the direction of walking and for an impact on the foot progression angle.
Regarding displacements of individual markers, displacements of the thigh and tibia wands and the knee marker in the anterior–posterior axis had the largest calculated impact on kinematics, all with an RMSD angle of over 5° in the transversal plane (Figure 4). These findings confirmed previous results demonstrating the knee marker’s high impact in the anterior–posterior axis in the transversal plane but it’s very low impact when displaced in the proximal–distal axis12,13. Even though some studies have reported improvements in calibration methods, such as the Knee Alignment Device, marker placement reproducibility and reliability remains the CGM’s most significant limitation17. The CGM’s high sensitivity to wand orientation is even more critical as the lack of an anatomical landmark makes its placement somewhat subjective. Current user manual specifications for wand placement are simply, “Adjust the position of the marker so that it lies in the plane that contains the hip and knee-joint centers and the knee flexion/extension axis.”18.
The CGM is characterized by a hierarchical, anatomical, top-down approach; therefore, a displaced marker affects the kinematics of every joint located distally to the anatomical segment containing that marker and the joint most proximal to it15. Additionally, the slight impact that we calculated on the foot progression angle demonstrates that without the medial markers of the knee and ankle, defining the joint centers is affected by multiple marker displacements. Thus, an error in the placement of the knee-joint center marker impacts the definition of the ankle-joint center and consequently the foot progression angle. Overall, the calculated impact of a displaced marker could be noted in the two simulated displacements in opposite anatomical directions.
Gait scores, like the GPS, are very good at classifying a patient’s gait by comparing it with a reference database of a general asymptomatic population. As the calculation uses kinematic data, the variability noted because of marker displacement also introduces variability into the final gait classification and thus may also have a considerable impact on gait data interpretation. We therefore investigated the impact of marker displacement on overall gait scores. Marker displacement in one leg resulted in GPS variations of up to 7°. This is comparatively much greater than the 1.6° rated as the minimal variation of clinical significance19. As the GPS is calculated using the kinematics of both lower limbs, the variation expected if our simulations were applied to both sides would be even higher.
The impact of the variability of marker placement on our simulated gait kinematics is shown in Figure 5 by the corridors of maximal RMSD angle calculated per frame in the gait cycle added around one subject’s original curve. We note that the error can be defined by an overall offset added to the original data. This finding agrees with previous results reporting that the impact of errors on axis definition was more like an offset to the kinematics than a change in their overall pattern4. Such results may be useful for estimating the expected variability in kinematics when considering expected marker placement variability. To evaluate the impact of marker misplacement more accurately, our results could be used in combination with those from studies reporting on the precision of marker placement, such as Della Croce et al.11. Thus, the magnitude of each marker’s misplacement would be defined based on experimentally observed error.
The present study had some limitations. Firstly, the lack of literature regarding gait analysis’ sensitivity to marker placement makes comparisons with our results difficult. Secondly, marker displacement was done virtually, so the effects of soft tissue artefacts could not be considered. Different marker displacement distances and axes could also induce different soft tissue artefacts20,21. Moreover, our reference marker placements cannot be considered as ‘true’ references as they too were subject to the uncertainty of marker placement. We only applied displacements of 10 mm in only four directions, although that distance was defined according to Della Croce’s results and to serve as a potential standard reference for future comparisons11. Finally, the enormous amount of simulations required to compute every potential combination of marker displacement for the twenty subjects required enormous computing time. This imposed limits on the testing of numerous displacement distances and directions, as previously reported for single-marker displacements13.
To conclude, we performed a very extensive sensitivity analysis combining 390,625 simulated marker placements. We successfully identified the most sensitive angles contributing to an overall marker displacement simulation measurement and quantified the RMSD angles associated with the displacements of the different lower-limb markers. We also identified and analyzed simulated worst-case marker displacement scenarios. Additionally, we reported on which markers and which axes caused the greatest variability in the angles measured. Greater accuracy in the placement of thigh and tibia wands (or markers) and lateral femoral epicondyle markers in the anterior–posterior axis are required to improve the reliability of gait analysis using the GCM.